Aerial or satellite imagery can provide a bird's eye view of swaths of land. Multi-spectral images capture information beyond the traditional visible color bands into other spectrums such as infrared or thermal spectrums. The images may be analyzed by human experts to glean important information about the area imaged. The results of such analysis may be used to drive informed decisions as relevant for different applications. For example, a forestry or land-management organization may use aerial images to help identify de-forestation rates. Similarly, farmers may use aerial images to inspect the health of crops through the season and make watering and fertilization decisions for the crops at a daily or weekly cadence. Mapping companies may use aerial images to identify roads, new construction, or changes in the flow of traffic to provide more accurate maps.
Often, acquired images are not accurately mapped to the specific geographical locations. Despite the presence of global positioning system (GPS) information within the image, the image may not align with the known geographical structures such as buildings, farms, roads, or natural landmarks with precise GPS coordinates. Such inaccuracies in location may be due to errors in the image acquisition process, or during a stitching process where multiple images are stitched together into a single larger image. Human experts have to manually correct the errors in alignment in a slow and tedious manual process. The manual process can also be error prone, with different people aligning the images differently. The problem is exacerbated when a large number of images are acquired at a regular cadence to cover large regions.